harness-architecture-rules

Pass

Audited by Gen Agent Trust Hub on Jul 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes the command npx @nextstage-brasil/harness sync to regenerate adapters and synchronize project-specific rules within the harness directory.
  • [EXTERNAL_DOWNLOADS]: The skill fetches and runs the @nextstage-brasil/harness package from the NPM registry via npx to perform rule synchronization.
  • [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it processes untrusted data from the local repository to generate rules.
  • Ingestion points: The agent scans project manifests (such as package.json, composer.json, pyproject.toml, go.mod, Cargo.toml, and docker-compose.yml), directory trees, and existing documentation files (README.md, docs/).
  • Boundary markers: There are no instructions to use specific delimiters or include warnings that the agent should ignore instructions embedded within the content of the scanned files.
  • Capability inventory: The agent is authorized to write new files to the repository (architecture-rules.md) and execute shell commands through the vendor's CLI tool.
  • Sanitization: The skill does not define any validation or sanitization procedures for the metadata or text extracted from scanned files before incorporating them into the final AI instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 24, 2026, 08:16 PM
Security Audit — agent-trust-hub — harness-architecture-rules